Search

This paper focuses on the clustering segmentation of 3D color point cloud. We extend the mean shift algorithm to the 3D xyz space, and what's more, we also consider the rgb color information, so the 6 dimensional data is adopted in the algorithm. The cluster center converges to the joint position of the local maximum density and the minimum gradient change of color, so our clustering segmentation not only considers the local geometrical features, but also utilizes the color information. The experiments show that our segmentation has better region consistency and has clear segmenting border in different color neighbors.

The package and system level temperature and thermal stress distributions of 10 W light emitting diode (LED) with 4 chips and 100 W LED with 100 chips were investigated using finite element analysis. The chips were arranged on a Si sheet which is soldered on the copper/diamond composite slug with very high conductivity. The experimental results show that the maximal temperature appears in the chips of both two high power LEDs packages. Compared with the 10 W LEDs package with 4 chips array, the heat issue caused by stacking and coupling of the heat in 100 W LEDs package with 100 chips array is more serious. The chip temperature in the center of the array is much higher, and it decreases with the distance between the chip and the center of LEDs increases. Great thermal stress lies between the chips and the solder, which will reduce the reliability of the package.

Hydrogen is a clean energy carrier. Converting abundant coal sources and green biomass energy into hydrogen effectively and without any pollution promotes environmental protection. The co-gasification performance of coal and a model compound of biomass, carboxymethylcellulose (CMC) in supercritical water (SCW), were investigated experimentally. The influences of temperature, pressure and concentration on hydrogen production from co-gasification of coal and CMC in SCW under the given conditions (20–25 MPa, 650°C, 15–30 s) are discussed in detail. The experimental results show that H2, CO2 and CH4 are the main gas products, and the molar fraction of hydrogen reaches in excess of 60%. The higher pressure and higher CMC content facilitate hydrogen production; production is decreased remarkably given a longer residence time.

This paper extends mean shift algorithm to 3D color point cloud, and successfully realizes its clustering segmentation. Then with these segmentation blocks, we use Random Sample Consensus (RANSAC) algorithm to calculate each block's approximate plane, and keep the points which are the nearest to the plane, thus the point cloud is simplified. The experiment shows that our simplifying results are perfect.

This paper gives a simplifying way of color point cloud. Firstly, we get the raw point cloud from the faro 3D Laser scanner, then we design our data structure to fit for managing the color point cloud, thirdly, utilizing statistical way we efficiently remove the outlier points and remain the inliers points. In the end, the point cloud is simplified based on voxels. Experiment shows that our method can directly manage the point cloud, and make efficient compression to the point cloud. At the same time, it can keep the color information, and the feature information is also well kept.

This paper proposes the normal vector estimation algorithm based on KD-Tree. The speed of searching neighbor field is improved by utilizing KD-Tree data structure. The orientation of normal vectors computed by PCA is ambiguous and confused. In order to solve this problem, the viewpoint of point cloud is set to check and flip over the orientation. For the purpose of obtaining the correct normal vector estimation, the scope of neighbor field is extended to improve the antinoise ability of normal vector estimation. The experiment result proves that the normal vector estimation algorithm is robust.

This paper proposes the point feature histogram based on the correct normal vector estimation. The four dimensional features of each point in point cloud is computed by synthesizing the normal vector information of neighbour field of point cloud. All of four features are binned into histogram. The different type geometric primitives (such as plane, sphere, cylinder etc.) are generated to analyze the points' signature, and algorithm complexity is reduced by approximating factor parameter. The experiment result proves that point feature histogram has the discriminative power.